General questions

Course Q&A Deep Learning Specialization Neural Networks and Deep Learning
1

Hi! After the DLS course 1, I have some general questions as follows:

  1. We are only using Linear Regression in course 1, all the Z’s are in a linear relationship with the feature X. Are we going to learn and implement Polynomial Regression or other regression in later courses?

  2. Why are we using activation functions for each hidden layer unit? I know that from the course, professor said that we are just duplicating what we had done in logistic regression model. But will it change things a lot for the situation where we only use, say, one activation function in the output layer?

  3. I am current studying Professor Andrew Ng’s Machine Learning course. The terminologies and concepts are quite similar comparing to DLS course. What are the main differences between machine learning and deep learning in terms of algorithms, methodologies, or anything else?

Thanks a lot!

2

Here are some responses by question:

  1. No, we only use linear activation in each layer, followed by a non-linear activation function. There is no “polynomial regression” as Prof Ng describes in Week 2 of the Stanford ML course. But notice that is a “pre-processing step” that you could try with any of the network architectures we have here in this course.

  2. I think you missed Prof Ng’s point about activation functions. It is essential that we include a non-linear activation function at every layer in a Neural Network. If you do not do this, then there is literally no point in having a multi-layer network. That is because it is an easily provable theorem that the composition of linear functions is still a linear function.

  3. Machine Learning is a more general concept than Deep Learning. Note that the Prof Ng’s Machine Learning course was created quite a while ago (2011), so the Neural Networks section of that course (weeks 3 and 4) uses an older form of notation and things are somewhat different. ML includes lots of other types of algorithms that are not covered here in DL, e.g. Anomaly Detection, Linear Regression, Spam Detection, Unsupervised Learning, Recommender Systems. It’s good to know about those other types of algorithms, so the Stanford Machine Learning course is a valuable thing to take in addition to these courses.

3 Likes
3

Thanks for answering. May I ask where did Prof Ng mentioned there is no polynomial regression? I must missed it during the lecture.

4

I just meant that he does not introduce or cover that topic here in the Deep Learning Specialization. It was covered in Week 2 of the Stanford Machine Learning course.